A Study on the Effects of Sympathetic Skin Response Parameters in Diagnosis of Fibromyalgia Using Artificial Neural Networks SYSTEMS-LEVEL QUALITY IMPROVEMENT A Study on the Effects of Sympathetic Skin Response Parameters in Diagnosis of Fibromyalgia Using Artificial Neural Networks Ozhan Ozkan1 & Murat Yildiz1 & Evren Arslan1 & Sedat Yildiz2 & Suleyman Bilgin3 & Selami Akkus4 & Hasan R. Koyuncuoglu5 & Etem Koklukaya6 Received: 22 April 2015 /Accepted: 16 November 2015 /Published online: 8 December 2015 # Springer Science+Business Media New York 2015 Abstract Fibromyalgia syndrome (FMS), usually observed commonly in females over age 30, is a rheumatic disease accompanied by extensive chronic pain. In the diagnosis of the disease non-objective psychological tests and physiologi- cal tests and laboratory test results are evaluated and clinical experiences stand out. However, these tests are insufficient in differentiating FMS with similar diseases that demonstrate symptoms of extensive pain. Thus, objective tests that would help the diagnosis are needed. This study analyzes the effect of sympathetic skin response (SSR) parameters on the auxil- iary tests used in FMS diagnosis, the laboratory tests and physiological tests. The study was conducted in Suleyman Demirel University, Faculty of Medicine, Physical Medicine and Rehabilitation Clinic in Turkey with 60 patients diag- nosed with FMS for the first time and a control group of 30 healthy individuals. In the study all participants underwent laboratory tests (blood tests), certain physiological tests (pul- sation, skin temperature, respiration) and SSR measurements. The test data and SSR parameters obtained were classified using artificial neural network (ANN). Finally, in the ANN framework, where only laboratory and physiological test re- sults were used as input, a simulation result of 96.51 % was obtained, which demonstrated diagnostic accuracy. This data, with the addition of SSR parameter values obtained increased to 97.67 %. This result including SSR parameters –meaning a higher diagnostic accuracy – demonstrated that SSR could be a new auxillary diagnostic method that could be used in the diagnosis of FMS. Keywords Biomedical signal processing . Biomedical data analysis . Fibromyalgia syndrome (FMS) . Sympathetic skin response(SSR) .Autonomicnervoussystem(ANS) .Artificial neural networks (ANN) Introduction Fibromyalgia syndrome (FMS) is a common rheumato- logic syndrome which is manifested as chronic and dif- fuse musculoskeletal pain. When examined closely, an early diagnosis is possible between the ages of 29–37, however due to the difficulties in diagnosis and the neglect of the disease by the patients the age of diagnosis rises to the range of 34–53 [1]. 70–90 % of all FMS patients are women [1]. Thus, it could be stated that FMS is prevalent in women over the age of 30. As a result of studies conducted by the American College of Rheumatology (ACR) in 1990’s, it was observed that in FMS patients there was a history of extensive pain for more than three months in at least 11 of the 18 sore points selected based on the condition of pain in muscular tissues. Thus, the This article is part of the Topical Collection on Systems-Level Quality Improvement. * Evren Arslan earslan@sakarya.edu.tr 1 Department of Electrical and Electronics Engineering, Faculty of Engineering, Sakarya University, 54187 Sakarya, Turkey 2 Physical Medicine and Rehabilitation Clinic, Egirdir Bone & Joint Disease Treatment and Rehabilitation Hospital, Isparta, Turkey 3 Department of Electrical and Electronics Engineering, Faculty of Engineering, Akdeniz University, Antalya, Turkey 4 Department of Physical Medicine and Rehabilitation, Faculty of Medicine, Yildirim Beyazit University, Ankara, Turkey 5 Department of Neurology, Faculty of Medicine, Suleyman Demirel University, Isparta, Turkey 6 Department of Electrical and Electronics Engineering, Faculty of Engineering, Gazi University, Ankara, Turkey J Med Syst (2016) 40: 54 DOI 10.1007/s10916-015-0406-0 http://orcid.org/0000-0003-2635-2605 http://crossmark.crossref.org/dialog/?doi=10.1007/s10916-015-0406-0&domain=pdf most important issue to be considered in FMS diagnosis is the existence of extensive pain in at least 11 of these 18 points for more than three months [2]. Consequently, if the pain spots of the patients are the same as the points determined by ACR, further evaluations including psy- chological tests are conducted by the specialist physi- cians on the patient. On psychological tests, the severity equivalents of the pain spots were determined and if these are at an acceptable level, the diagnostic process is taken further using laboratory tests and physiological tests (auxiliary tests). Auxiliary tests are the final tests in evaluation of FMS diagnosis. Although these exhaustive tests are sufficient in the diagnosis of FMS, they are far from being sufficient in differentiation of FMS from similar diseases with extensive pain. Thus, it is necessary to investigate additional diagnostic methods in FMS diagnosis and exclusion of similar diseases. In 2010, a new approach was developed by ACR to measure pre-diagnostic criteria and severity of symptoms, which considered widespread pain in- dex and categorical scales for cognitive symptoms, unrefreshed sleep, fatigue, and a number of somatic symp- toms. The new approach ignored the tender point examina- tion, which was among the 1990 ACR criteria [2]. Sympathetic skin response (SSR), on the other hand, is an electrical potential recorded by a measurement device after an unexpected audible or an external electrical stimulus was applied on the electrical charge of an area of skin and reflects sudomotor neural activity stimulating sweat glands. Previous studies were not able to determine whether SSR data could be among the methods, which could be used, in direct diagnosis of FMS, however it was considered that it could be related to FMS since it responds to the dysfunc- tions in the autonomic nervous system (ANS) [3, 4]. A study that scrutinized the effects of SSR on FMS, the support of SSR parameters on the psychological tests, which are the major tests used for the diagnosis of the disease, was inves- tigated. It was observed that the accuracy of the diagnosis as a result of ANN analysis, where the psychological test re- sults were evaluated together with the SSR parameters, was higher than the case where only the psychological test re- sults were considered for the diagnosis [5]. In a study by Ulas et al. conducted to investigate SSR changes in FMS, it was determined that the response latency parameter on SSR signal recorded on the palm and soles of the patients was longer than healthy individuals. The results of that study demonstrated that the changes in the autonomic nervous system due to the disease could be graded by SSR measure- ments and by using soft computing methods at the end of the research, more significant information could be obtained from the signals [4]. Ozgocmen et al. studied the relation- ship between SSR response latency parameter and the Hamilton Anxiety Test (HAM-A) that is one of the psycho- logical test utilized for FMS diagnosis, and stated that these two parameters could be correlated at the end of the study [6]. In another study, based on the information that FMS causes dysfunctions in the autonomic nervous system, Tervainen et al. investigated the relationship between certain parameters of SSR and autonomic nervous system [7, 8]. Everhart and Harrison scrutinized the relationship between SSR measurements and their functions in the autonomic nervous system and their findings demonstrated that auto- nomic nervous system could be graded by SSR measure- ments [9]. A study by Ahuja et al. considered that SSR signal analysis could become more analytical with soft com- puting methods and they determined that a new test method could be proposed to help with the diagnosis and treatment of psychological and psychiatric diseases that affect the au- tonomic nervous system [10]. In another study that exam- ined the effects of SSR on FMS, laboratory tests, one of the auxiliary methods used in the diagnosis of the disease, were administered to the patients and with the aid of a measure- ment system placed in a special test environment, SSR pa- rameter values were obtained. It was observed by consider- ing the simulation results obtained as a result of ANN anal- ysis of these data that laboratory tests were supported by SSR and increased simulation accuracy [11]. Berger et al. investigated the relations between the SSR and autonomic dysfunction in spinal cord injury (SCI) and they concluded that the SSR can be used together with the other validated autonomic tests in the evaluation of SCI [12]. In another study, Saba and Sultan examined the changes in the autonomic nervous system for the rheumatoid arthritis (RA) disease. SSR values of the patients and the con- trols were measured to evaluate the changes in ANS for RA. They inferentially presented that the statically sig- nificant relation between the autonomic nervous system and the SSR [13]. In another study by Dag et al. studied on autonomic dysfunctions in patients with polycystic ovary syn- drome. They concluded that the polycystic ovary syndrome caused dysfunctions in the autonomic nervous system and this effect was showed easily by SSR and R-R interval tests [14]. In this study, based on the assumption that FMS affects autonomic nervous system and SSR in turn is affected by the dysfunctions in the autonomic nervous system, the effects of SSR on the auxiliary tests applied in FMS diagnosis were examined in detail. In this context, physiological tests and laboratory tests were conducted on patients diagnosed with FSM and healthy individuals (control group) se- lected by specialist physicians in Suleyman Demirel University, Faculty of Medicine, Physical Medicine and Rehabilitation Clinic, and SSR measurements were ob- tained from all participants. For the analyses, a multi- layer feed forward neural network (MLFFNN) was cre- ated and used to scrutinize the effects of the SSR parameters calculated based on SSR measurements on the diagnosis of the disease. 54 Page 2 of 9 J Med Syst (2016) 40: 54 Materials and Methods Data Acquisition and Parameter Computation The patients, included in the study after the ethical committee approval, were selected from the patients that were diagnosed with FMS for the first time and applied to Suleyman Demirel University, Faculty of Medicine, Physical Medicine and Rehabilitation Clinic. The diagnoses were based on the criteria stipulated by ACR in 1990. The control group of the study was formed by healthy and volunteer individuals who have applied to the outpatient clinics of the same hospital for routine check- ups or were hospital employees, paired with the patient groups with respect to age and gender. Those undergoing a treatment that would affect SSR values, with FMS coexisting diseases, taken alcohol, tobacco, tea or coffee during the last twenty-four hours before the measurement, consumed a meal with high calories during the last six hours before the measurement, not willing to obey measurement protocols were excluded from the study. The patients and healthy individuals were examined by specialist physicians for conformity to the established criteria. The participants were informed about the mea- surement process and they have signed written assent forms. Then, laboratory tests were performed and after 15 min of rest, heart and respiration rates per minute were taken. Before this process, the skin temperatures of the area that SSR measurements will be taken from were determined and average skin temperatures were calculated. SSR records were taken in a silent and illuminated room with a suitable temperature (22 °C – 24 °C), while the patient was in a sitting position using a system set up at Physical Medicine and Rehabilitation Clinic (PowerLab 8/30 with LabChart Pro, Data Acquisition Systems, AD Instruments, Inc. Colorado Springs, CO, USA). Ring electrodes placed at the middle bone level on the left hand second and third fingers were used as recording electrodes. Stimulations were given using a stimulator placed at the median nerve block in the right forearm area. The stimulation current strength applied was 0.01A. To prevent the decrease in the reaction of the skin against the stimulation due to the repetition of the stimulation, the lag between two stimulations were adjusted to be more than 30 s and 5 stimulations were applied in ran- dom intervals. When the measurement quality was de- termined to be low, the number of stimulations was increased and when no response to at least 5 stimulations was observed the measurement was classified as Bno response^ and excluded from the study. The parameters used in the studywere SSR response latency time (SSRLT), SSR maximum amplitude (SSRMAX-A), the time interval between the stimuli applied (SSRTT), laboratory test results (blood tests) and physiological test (heart rate, skin temperature, respiration measurements) results. SSR response latency time is the time that passes between the initiation of the stimulus artifact and the beginning point of the change in sym- pathetic skin response [15]. Maximum amplitude value is the Table 2 Minimum, maximum, mean and standard deviation values of CRP, RF, Sedim., WBC, HB, PLT values for all participants (a), patients (b) and controls (c) respectively) Min. Max. Mean STD. a. All Ss. (86) CRP 2 20 3,967 2,609 RF 8,060 29 10,177 3,273 Sedim. 1 65 15,860 11,204 WBC 3800 11900 7152,326 1463,897 HB 11,100 16,700 13,557 1,173 PLT 142000 387000 259279,070 52924,286 b. Patients (57) CRP 2,660 20 4,175 2,982 RF 8,060 29 10,109 3,156 Sedim. 1 65 16,351 12,392 WBC 3800 11900 7245,614 1544,121 HB 11,100 16,700 13,586 1,210 PLT 142000 387000 268315,789 57206,443 c. Controls (29) CRP 2 10,400 3,558 1,618 RF 9,220 28 10,312 3,545 Sedim. 5 41 14,897 8,516 WBC 5000 10000 6968,966 1297,829 HB 11,600 16,500 13,500 1,116 PLT 153000 308000 241517,241 38273,098 Ss Subjects, Temp. Temperature, STD Standard deviation, RF Rheumatoid factor, CRP C-reactive protein, WBC White blood cell, HB Hemoglobin, PLT Platelet, Sedim. Sedimentation Table 1 Minimum, maximum, mean and standard deviation values of SSR parameters for all subjects (a), patients (b) and controls (c) respectively Min. Max. Mean STD. a. All Ss. (86) LT (sn) 0,365 17,246 2,478 3,148 TT (sn) 28,250 70,610 46,888 9,259 Max-A (μV) 0,005 9,140 2,760 2,282 b. Patients (57) LT (sn) 0,417 17,246 2,902 3,547 TT (sn) 28,250 67,644 47,036 8,562 Max-A (μV) 0,005 9,140 2,680 2,267 c. Controls (29) LT (sn) 0,365 3,942 1,328 0,926 TT (sn) 28,288 70,610 46,488 10,918 Max-A (μV) 0,258 8,810 2,950 2,312 Ss Subjects, LT Latency time, TT Total time, Max-A Maximum ampli- tude, STD Standard deviation, SSR Sympathetic skin response J Med Syst (2016) 40: 54 Page 3 of 9 54 largest potential change obtained between two stimuli [16]; and the time between the stimuli is the time that passes between a stimulus applied to measure the SSR and the successive stim- ulus that follows. The numerical values of these parameters are obtained using calculations on Matlab (http://www.mathworks. com/products/matlab/) mathematical programming software using SSR graphics transferred into computing environment via the interface on the measurement system. Minimum, maximum, mean and standard deviation values for the SSR parameter values obtained from the patient and control groups are displayed in Table 1. One of the auxiliary tests conducted in laboratory environ- ment, blood tests are rheumatoid factor (RF), c-reactive pro- tein (CRP), white blood cell count (WBC), hemoglobin count (HB), platelet count (PLT) and sedimentation (sedim) mea- surements conducted to differentiate the disease with other diseases with similar symptoms [17]. These tests were imple- mented by specialist physicians and recorded. Other auxiliary tests are the physiological tests; namely the skin temperature, heart rate and respiratory rate. Skin temperature of the partic- ipants were taken by specialist physicians at Suleyman Demirel University, Faculty of Medicine, Physical Medicine and Rehabilitation Clinic, using the skin temperature probe of the Cadwell Sierra Wedge EMG/NCV brand, 2 channel EMG device. Per minute hear rates of the participants were deter- mined by the examination and SSR records of the participants. Per minute respiratory rate of the participants was determined by examination. Minimum, maximum, mean and standard deviation values of blood tests obtained from the patient and control groups are displayed in Table 2. Figure 1 and Fig. 2 demonstrates the graphs on these phys- iological test results obtained from the patients with FMS and the control group, Table 3 displays the minimum, maximum, men and standard deviation valued for these. These two fig- ures display the number of participants for control group and patient group as 30 and 60 respectively; however the number of participants that measurements could be taken was 29 for the control group and 57 for the patient group. No responses were obtained from 3 patients and 1 healthy individual for at least 5 stimuli and hence they were excluded from the system. Thus, the total number of participants who were excluded from the study was 86. Table 3 Minimum, maximum, mean and standard deviation values of skin temperature, pulsation and respiration values for all participants (a), patients (b) and controls (c) respectively Min. Max. Mean STD. a. All Ss. (86) Skin Temp. 30 34 32,912 0,782 Pulsation 60 120 80,278 11,087 Respiration 12 27 17,789 2,953 b. Patients (57) Skin Temp. 30 34 33,012 0,642 Pulsation 60 120 80,767 11,585 Respiration 13 27 18,033 2,869 c. Controls (29) Skin Temp. 30 34 32,713 0,976 Pulsation 63 110 79,300 9,944 Respiration 12 25 17,300 3,057 Ss Subjects, Temp. Temperature, STD Standard deviation Fig. 2 Physiological test results for patients (60 subjects) Fig. 1 Physiological test results for controls (30 subjects) 54 Page 4 of 9 J Med Syst (2016) 40: 54 http://www.mathworks.com/products/matlab/ http://www.mathworks.com/products/matlab/ In the study, where the classifications were conducted using the MLFFNN method, detailed in the next section, 7 different input layers that use different combinations of phys- iological tests, which were heart rate, skin temperature and respiratory rate, laboratory tests (blood tests), and SSR param- eters, which were response latency time, total time and max- imum amplitude values, were designed. These input layers were the following: 1. Laboratory test results and physiological test results 2. Physiological test results and SSR parameters 3. Laboratory test results, physiological test results and SSR parameters 4. Physiological test results 5. Laboratory test results 6. SSR parameters 7. Laboratory test results and SSR parameters The last four cases, where only physiological test results, only laboratory test results, only SSR parameters and SSR parameters and laboratory test results together were used, were scrutinized in a previous study [11]. But we performed again the all cases with k-fold cross validation (CV) method for an accurate comparison. So this study evaluates these four cases and the three newly examined three cases together. Classification results are detailed in the next section. Multilayer Feedforward Neural Network (MLFFNN) One of the most common methods used to solve various prob- lems such as information classification and evaluation today is artificial neural networks. ANN are information processing sys- tems developed based on biological neural networks and the main objective in an ANN approach is to learn about an event based on the way human brain functions [18]. There are several different frameworks developed for ANN. One of these frame- works is the multilayer feedforward neural network. In this net- work, information only moves forward secretly and towards the output layer. Signals are always transported forward and con- nection weights could be changed during training [19]. Generally, feedforward neural networks are divided into two types; single-layer and multilayer neural networks. If the net- work structure designed to analyze a specific problem consists of only input and output layers, this structure is a single-layer feedforward network framework. In addition to the input and output layers in a single-layer networks, if a network contains one or more hidden layers, it is called a multilayer feedforward network. A sample feedforward network structure is displayed in Fig. 3. The layers in the network are expressed as the input, hidden, and output layers respectively. In transmission of the data in the input layer to the neurons in the hidden layer, the neurons in the input layer are utilized. The neurons in the hidden layer utilize the output of the preceding layer as input [20]. In the figure, the weights between the input layer and the hidden layer are indicated as W1,ji, the weights between the hidden layer and the output layer are indicated as W2,yj. i = 1,2,... n are input neurons, j = 1,2,...,m are hidden layer neurons and y = 1,2,...,k are output layer neurons [21, 22]. Since MLFNN is commonly used in similar studies in literature, this network structure was preferred in this study; the network was designed as multilayered since more reliable test results are observed when more than one hidden layers are utilized. Nh=5Ni=9 CRP RF HB Sedim WBC PLT Pulsation Skin Temp. Respiration 0=FMS 1=Healty Hidden Layer Output LayerInput Layer LW 1,1 LW 2,1 b 2b 1 11 No=1 CRP: C-reactive protein, RF: Rheumatoid factor, HB: Hemoglobin count, Sedim: Sedimentation, WBC: White blood cell count, PLT: Platelet count, Temp: Temperature Fig. 4 ANN model, laboratory and physiological test scores are inputs. SSR: Sympathetic skin response, SSRLT: SSR latency time, SSRTT: SSR total time, SSRMaxA: SSR maximum amplitude, Temp: Temperature Fig. 3 Multilayer feedforward neural network model. CRP: C-reactive protein, RF: Rheumatoid factor, HB: Hemoglobin count, Sedim: Sedimentation, WBC: White blood cell count, PLT: Platelet count, Temp: Temperature J Med Syst (2016) 40: 54 Page 5 of 9 54 Simulation Results and Discussion In this study, classifications were conducted using multilayer feedforward network models, which were mentioned briefly in the previous section and of which input and output are displayed in Figs. 4, 5 and 6. All analyses are performed in MATLAB (http://www.mathworks.com/products/matlab/) neural networks tool. In the study, initially physiological and laboratory test results were analyzed with ANN. For this purpose a multilayer feedforward ANN structure with an input layer composed of 9 neurons, a hidden layer of 5 neurons and an output layer of 1 neuron was designed (Fig. 4). Secondly, the effects of SSR data on physiological tests were examined. For this purpose a multilayer feedforward ANN structure with an input layer composed of 6 neurons, a hidden layer of 8 neurons and an output layer of 1 neuron was used (Fig. 5). Finally, the analyses were completed by adding laboratory test results to the physiological test results and SSR parameters. The multilayer feedforward artificial neural net- work structure designed in this step had an input layer with 12 neurons, a hidden level with 9 neurons and an output layer with 1 neuron (Fig. 6). In the output level, individuals with disease were identified with a B0^ and healthy individ- uals were identified with a B1^. In the hidden layer of the network designed, 30*48 neurons were utilized. In the analysis of the test findings, the number of neurons that yielded 100 % accurate training during the training pro- cess and with the highest simulation accuracy was used. We used k-fold CV to reduce the bias related with random sampling of dataset. K-fold CV divvies dataset up among k different subsets that have almost the same size. We examined different k values and chose four for our dataset. The training and test set of ANNwere formed by the data collected from 86 participants (57 FMS, 29 healthy controls). During the evalu- ation of physiological test results with the laboratory test re- sults, the ANN trained with a 100 % accuracy. ANN classified the patients and healthy individuals and recorded them as patients by assigning a B0^ and as healthy by assigning a B1^ in the output level. As a result, in the evaluation of phys- iological test results with the laboratory test results, a simula- tion accuracy of 96.51 % was observed. When the physiolog- ical tests were supported with the data from SSR parameters, simulation accuracy was calculated as 95.35 % in total. In the final stage of the study, the physiological test results and the laboratory test results were evaluated with the inclusion of SSR parameter findings and accuracy rates (sensitivity) of 100 % for the patient group, (specificity) 93.10 % for the control group, and 97.67 % for the patient and control groups Nh=9Ni=12 SSRLT SSRMaxA SSRTT CRP RF HB Sedim WBC PLT Pulsation Skin Temp. Respiration 0=FMS 1=Healty Hidden Layer Output LayerInput Layer LW 1,1 LW 2,1 b 2b 1 11 No=1 SSR: Sympathetic skin response, SSRLT: SSR latency time, SSRTT: SSR total time, SSRMaxA: SSR maximum amplitude, Temp: Temperature, CRP: C-reactive protein, RF: Rheumatoid factor, HB: Hemoglobin count, Sedim: Sedimentation, WBC: White blood cell count, PLT: Platelet count, Temp: Temperature Fig. 6 ANN model, SSR parameters, laboratory test scores, and physiological test scores are all inputs Nh=8Ni=6 SSRLT SSRMaxA SSRTT Pulsation Skin Temp. Respiration 0=FM 1=He Hidden Layer Output LayerInput Layer LW 1,1 LW 2,1 b 2b 1 11 No=1 SSR: Sympathetic skin response, SSRLT: SSR latency time, SSRTT: SSR total time, SSRMax SSR maximum amplitude, Temp: Temperature Fig. 5 ANN model, SSR parameters and physiological test scores are inputs. SSR: Sympathetic skin response, SSRLT: SSR latency time, SSRTT: SSR total time, SSRMaxA: SSR maximum amplitude, Temp: Temperature, CRP: C-reactive protein, RF: Rheumatoid factor, HB: Hemoglobin count, Sedim: Sedimentation, WBC: White blood cell count, PLT: Platelet count, Temp: Temperature 54 Page 6 of 9 J Med Syst (2016) 40: 54 http://www.mathworks.com/products/matlab/ in total were concluded (overall accuracy). The simulation accuracy rates for these studies are displayed in Tables 4. We also reanalyzed the cases (first four) with k-fold CVmeth- od in our previous study for an accurate comparison. Table 4 also demonstrates the simulation accuracies of these cases. Some other statistical parameters are available in Table 5. Table 6 displays discriminate analysis and t-test results con- ducted using SPSS (Statistical Package for Social Sciences) for all groups. Furthermore Table 6 displays the simulation results of the previous related study dem- onstrating that the simulation accuracy increases when laboratory tests used in FMS diagnosis are supported with SSR parameters [11]. Figure 7 displays the flow chart for the study and a summary of total simulation accuracies for all cases together. Our previous study demonstrated that SSR supported the laboratory test results in the diagnosis of FMS via the simula- tion accuracy rates obtained using ANN [11]. This study an- alyzed the effects of SSR on auxiliary laboratory tests and the physiological tests in the diagnosis of FMS using ANN. For example, the simulation accuracy as a result of the classifica- tion of laboratory and physiological tests using ANN was calculated as 96.51 %. When it is considered that the simulation accuracy of the laboratory tests on their own was 95.35 % and the simulation accuracy of the physiological tests on their own was 95.35 %, it was determined that the evalu- ation of physiological test findings together with laboratory tests increased the simulation accuracy. On the other hand, when physiological test results were supported by SSR find- ings an accuracy of 95.35 % was obtained. The last step of the study, the simulation accuracy of 97.67 % obtained as a result of the classification of the laboratory test, physiological test and SSR data using ANN is an indicator of the support SSR data provides for both laboratory and physiological tests in FMS diagnosis. Furthermore, the simulation accuracy of 86.4 % (97.67 % with k-fold CV in recent analysis) that was found in the previous study where the laboratory tests were supported with SSR findings demonstrated that SSR parame- ters supported the laboratory tests more than the physiological tests [11]. However, it was also observed that all auxiliary methods had different contributions in the diagnosis of the disease. For instance, when the physiological tests were eval- uated with the laboratory tests, the sensitivity for the patient group is same, but specificity has risen to 96.55 % in the differentiation of healthy individuals. When SSR param- eters were added to these tests, the evaluation showed Table 5 Statistical parameters for overall subjects Phys. T. Lab. T. SSR Lab. T. &SSR Lab. T. &Phys. T. Phys. T. &SSR Lab. T. &SSR &Phys. T. OVERALL PLR 13,99 13,99 9,33 14,50 27,98 13,99 14,50 NLR 0,04 0,04 0,04 0,00 0,04 0,04 0,00 PPV 0,96 0,96 0,95 0,97 0,98 0,96 0,97 NPV 0,93 0,93 0,93 1,00 0,93 0,93 1,00 Lab. T. Laboratory tests, Phys. T Physiological tests, SSR Sympathetic skin response, PLR Positive likelihood ratio, NLRNegative likelihood ratio, PPV Positive predictive value, NPV Positive predictive value Table 4 Training, test and overall statistics (%) Phys. T. Lab. T. SSR Lab. T. &SSR Lab. T. &Phys. T. Phys. T. &SSR Lab. T. &SSR &Phys. T. TRAINING Specificity 100 100 100 100 100 100 100 Sensitivity 100 100 100 100 100 100 100 Accuracy 100 100 100 100 100 100 100 TEST Specificity 71,43 71,43 57,14 71,43 85,71 71,43 71,43 Sensitivity 86,67 86,67 86,67 100 86,67 86,67 100 Accuracy 81,82 81,82 77,27 90,91 86,36 81,82 90,91 OVERALL Specificity 93,10 93,10 89,66 93,10 96,55 93,10 93,10 Sensitivity 96,49 96,49 96,49 100 96,49 96,49 100 Accuracy 95,35 95,35 94,19 97,67 96,51 95,35 97,67 Lab. T. Laboratory tests, Phys. T Physiological tests, SSR Sympathetic skin response J Med Syst (2016) 40: 54 Page 7 of 9 54 that the sensitivity has risen to 100 %, while specificity for the differentiation of healthy individuals has fallen to 93.1 %. Furthermore, according to the results of the t-test conducted using SPSS, since p value for SSRLT (SSR), PLT (Lab. Test) data were smaller than 0.05 and for skin temperature (Phys. Test) was smaller than 0.1, it could be stated that these three parameters were the most significant three parameters in differentiation of ill and healthy individuals. Thus, the discriminate analysis results for only these three parameters are in support of this argument (Table 6). When the results of other discriminate analysis are considered, it is observed that total accuracy increased when laboratory tests were supported by both SSR and physiologi- cal tests (Table 6). Conclusion As a result, it could be stated that evaluating more than one test all together more sufficient in FMS diagnosis and SSR Subjects Laboratory Tests SSR Signal Physiological Tests Maximum Amplitude Latency Time Total Time SSR Parameters Pulsation, Skin temp., Respiration Blood Tests Matlab ANN*ANN* ANNANN* ANN* ANN ANN 95.35%68.2% 95.35% 86.4% 94.19 96.51% 97.67% 95.35% 97.67% SSR: Sympathetic skin response, ANN: Artificial neural network Fig. 7 Basic flowchart of the study and overall simulation accuracies (with previous simulation). SSR: Sympathetic skin response, ANN: Artificial neural network Table 6 Discriminant analysis (a) and T test results (b) a. Phys. T. Lab. T. SSR Lab. T. &SSR Lab. T. &Phys. T. SSRLT&PLT & Skin Temp. Patients (57) 94,7 93 38,6 91,2 84,2 89,5 Controls (29) 10,3 3,4 75,9 20,7 34,5 34,5 All Ss. (86) 66,3 62,8 51,2 67,4 67,4 70,9 b. F Sig. CRP 2,617 0,11 PLT 5,307 0,02 Skin Temp. 3,068 0,08 SSRLT 13,68 0 Ss Subjects, Lab. T. Laboratory tests, Phys. T Physiological tests, SSR Sympathetic skin response, SSRLT SSR latency time,CRPC-reactive protein,PLT Platelet count, Temp. Temperature 54 Page 8 of 9 J Med Syst (2016) 40: 54 parameters are also used an auxiliary diagnosis method with laboratory and physiological tests. There is a significant con- tribution of the support of these tests by physiological tests measurements on more accurate differentiation of healthy in- dividuals and by SSR on more accurate differentiation of the individuals with disease. When the overall accuracy is consid- ered and when the laboratory tests and physiological tests were evaluated together, a simulation accuracy of 96.51 % was observed, while the laboratory tests were evaluated with the SSR parameters, the overall accuracy rose to 97.67 % and sensitivity rose to 100 %. This fact demonstrated the signifi- cance of SSR measurements in addition to laboratory tests, instead of physiological tests in differentiation of the patients from healthy individuals. However, due to the success of the physiological tests in differentiation of the healthy controls, it was proposed that, instead of exclusion of these tests, SSR measurements should be included among the auxiliary diag- nosis methods and all three techniques should be benefited from. On the other hand the results of this study showed that the SSR is an important parameter in assessment of FMS and FMS disease causes dysfunctions in the autonomic nervous system so that this supports the studies which examined the effect of the SSR on the autonomic nervous system. So this study also supports the relationship between the SSR and the FMS dysfunctions. Acknowledgments This research is funded by The Scientific and Technological Research Council of Turkey (TUBITAK) and Sakarya University (Project code: 2009-50-02-016) within the context of the pro- ject code: 108E036, and titled BEvaluation of HRV, SSR and Psychological Tests using Wavelet Transform and Artificial Neural Nets for the diagnosis of Fibromyalgia Syndrome and Determination of Relations^. Relevant reference number of ethic committee report is B.30.2.SDU.0.01.00.01.301.01/19-271. References 1. Boissevain, M.D., and McCain, G.A., Toward an integrated under- s tanding of f ibromyalgia syndrome. I . Medical and Pathophysiological Aspects. Pain. 45(3):227–238, 1991. doi:10. 1016/0304-3959(91)90047-2. 2. Wolfe, F., Clauw, D.J., Fitzcharles, M.A., Goldenberg, D.L., Katz, R.S., Mease, P., Russell, A.S., Russell, I.J., Winfield, J.B., and Yunus, M.B., The American college of rheumatology preliminary diagnostic criteria for fibromyalgia and measurement of symptom severity. Arthritis Care Res (Hoboken). 62(5):600–610, 2010. doi: 10.1002/acr.20140. 3. Ozkan, O., Yildiz, M., Bilgin, S., and Koklukaya, E., Measuring the sympathetic skin response on body and using as diagnosis- purposed for lung cancer patients by artificial neural networks. J Med Syst. 34(3):407–412, 2010. doi:10.1007/s10916-009-9253-1. 4. Ulas, U.H., Unlu, E., Hamamcioglu, K., Odabasi, Z., Cakci, A., and Vural, O., Dysautonomia in fibromyalgia syndrome: sympathetic skin responses and RR interval analysis. Rheumatol Int. 26(5):383– 387, 2006. doi:10.1007/s00296-005-0007-1. 5. Yildiz, M., Ozkan, O., and Koklukaya, E., Investigating the effect of sympathetic skin response parameters on the psychological test scores in patients with fibromyalgia syndrome by using ANNs. SAU J. Sci. 17(1):27–32, 2013. 6. Ozgocmen, S., Yoldas, T., Yigiter, R., Kaya, A., and Ardicoglu, O., R-R interval variation and sympathetic skin response in fibromyalgia. Arch Med Res. 37(5):630–634, 2006. doi:10.1016/j.arcmed.2005.11. 008. 7. Tarvainen, M.P., Koistinen, A.S., Valkonen-Korhonen, M., Partanen, J., and Karjalainen, P.A., Analysis of galvanic skin re- sponses with principal components and clustering techniques. IEEE Trans Biomed Eng. 48(10):1071–1079, 2001. 8. Tarvainen, M.P., Partanen, J., Karjalainen, P.A., Koistinen, A.S., and Valkonen-Korhonen, M., Principal component analysis of gal- vanic skin responses. Engineering in Medicine and Biology Society, 2000. Proceedings of the 22nd Annual International Conference of the IEEE 4:3011–3014, 2000. doi:10.1109/IEMBS. 2000.901513. 9. Everhart, D.E., and Harrison, D.W., Heart rate and fluency perfor- mance among high- and low-anxious men following autonomic stress. Int J Neurosci. 112(10):1149–1171, 2002. doi:10.1080/ 00207450290026120. 10. Ahuja ND, Agarwal AK, Mahajan NM, Mehta NH, Kapadia HN (2003) GSR and HRV: Its application in clinical diagnosis. Computer-Based Medical Systems, 2003. Proceedings. 16th IEEE Symposium 1:279–283. doi:10.1109/CBMS.2003.1212802 11. Ozkan, O., Yildiz, M., and Koklukaya, E., The correlation of labo- ratory tests and sympathetic skin response parameters by using artificial neural networks in fibromyalgia patients. J Med Syst. 36(3):1841–1848, 2012. doi:10.1007/s10916-010-9643-4. 12. Berger, M.J., Hubli, M., and Krassioukov, A.V., Sympathetic skin responses and autonomic dysfunction in spinal cord injury. J Neurotrauma. 31(18):1531–1539, 2014. doi:10.1089/neu.2014. 3373. 13. Saba, E., and Sultan, H., Autonomic nervous system changes asso- ciated with rheumatoid arthritis: clinical and electrophysiological study. The Egyptian Rheumatologist. 36(4):157–163, 2014. doi: 10.1016/j.ejr.2014.03.002. 14. Dag, Z.O., Alpua, M., Turkel, Y., and Isik, Y., Autonomic dysfunc- tion in patients with polycystic ovary syndrome. Taiwanese Journal of Obstetrics & Gynecology. 54(4):381–384, 2015. doi:10.1016/j. tjog.2015.03.002. 15. Ozkan, O., Yildiz, M., and Koklukaya, E., Improving diagnostic accuracy by supporting the laboratory tests which used for diagno- sis of fibromyalgia syndrome with the sympathetic skin response parameters. SAU J. Sci. 15(1):1–7, 2011. 16. Zhai, J., Barreto, A.B., Chin, C., and Li, C., Realization of stress detection using psychophysiological signals for improvement of human-computer interactions^. Proceedings of the IEEE Southeastcon. 1:415–420, 2005. doi:10.1109/SECON.2005. 1423280. 17. Evcik, D., Kizilay, B., and Gokcen, E., The effects of balneotherapy on fibromyalgia patients. Rheumatol Int. 22(2):56–59, 2002. doi: 10.1007/s00296-002-0189-8. 18. Oztemel, E., Yapay sinir aglari. Papatya Publishing, Istanbul, 2012. 19. Askin, D., Iskender, I., and Mamizadeh, A., Dry type transformer winding thermal analysis using different neural network methods. J. Fac. Eng. Arch. Gazi Univ. 26(4):905–913, 2011. 20. Altun, A.A., and Allahverdi, N., A new approach to recognition of fingerprints enhanced by filtering techniques with artificial neural networks. J. Fac. Eng. Arch. Gazi Univ. 22(2):227–236, 2007. 21. MATLAB® Documentation Neural Network Toolbox Help (2015) http://www.mathworks.com/help/nnet/index.html. Accessed 26 Feb. 2015 22. Oz C, Koker R, Cakar S (2002) Yapay Sinir Agları ile Karakter Tabanlı Plaka Tanıma. Elektrik – Elektronik – Bilgisayar Mühendisligi Sempozyumu (ELECO), Bursa J Med Syst (2016) 40: 54 Page 9 of 9 54 http://dx.doi.org/10.1016/0304-3959(91)90047-2 http://dx.doi.org/10.1016/0304-3959(91)90047-2 http://dx.doi.org/10.1002/acr.20140 http://dx.doi.org/10.1007/s10916-009-9253-1 http://dx.doi.org/10.1007/s00296-005-0007-1 http://dx.doi.org/10.1016/j.arcmed.2005.11.008 http://dx.doi.org/10.1016/j.arcmed.2005.11.008 http://dx.doi.org/10.1109/IEMBS.2000.901513 http://dx.doi.org/10.1109/IEMBS.2000.901513 http://dx.doi.org/10.1080/00207450290026120 http://dx.doi.org/10.1080/00207450290026120 http://dx.doi.org/10.1109/CBMS.2003.1212802 http://dx.doi.org/10.1007/s10916-010-9643-4 http://dx.doi.org/10.1089/neu.2014.3373 http://dx.doi.org/10.1089/neu.2014.3373 http://dx.doi.org/10.1016/j.ejr.2014.03.002 http://dx.doi.org/10.1016/j.tjog.2015.03.002 http://dx.doi.org/10.1016/j.tjog.2015.03.002 http://dx.doi.org/10.1109/SECON.2005.1423280 http://dx.doi.org/10.1109/SECON.2005.1423280 http://dx.doi.org/10.1007/s00296-002-0189-8 http://www.mathworks.com/help/nnet/index.html A Study on the Effects of Sympathetic Skin Response Parameters in Diagnosis of Fibromyalgia Using Artificial Neural Networks Abstract Introduction Materials and Methods Data Acquisition and Parameter Computation Multilayer Feedforward Neural Network (MLFFNN) Simulation Results and Discussion Conclusion References